# Exploring the efficacy of neural networks for trajectory compression and the inverse problem

arXiv: https://arxiv.org/abs/2107.08849 · HTML: https://mlsanity.com/paper/2107.08849/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 6 of its 10 references that are in the map.
- First posted on arXiv 19 Jul 2021.

## Builds on (most cited first, 6 of 6)

- [Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift](https://mlsanity.com/paper/1502.03167/) (2015)
- [Brax -- A Differentiable Physics Engine for Large Scale Rigid Body Simulation](https://mlsanity.com/paper/2106.13281/) (2021)
- [Improving Generalization Performance by Switching from Adam to SGD](https://mlsanity.com/paper/1712.07628/) (2017)
- [Survey of Deep Reinforcement Learning for Motion Planning of Autonomous Vehicles](https://mlsanity.com/paper/2001.11231/) (2020)
- [Towards a Robust Aerial Cinematography Platform: Localizing and Tracking Moving Targets in Unstructured Environments](https://mlsanity.com/paper/1904.02319/) (2019)
- [CocoNet: A deep neural network for mapping pixel coordinates to color values](https://mlsanity.com/paper/1805.11357/) (2018)

## Abstract (arXiv)

In this document, a neural network is employed in order to estimate the solution of the initial value problem in the context of non linear trajectories. Such trajectories can be subject to gravity, thrust, drag, centrifugal force, temperature, ambient air density and pressure. First, we generate a grid of trajectory points given a specified uniform density as a design parameter and then we investigate the performance of a neural network in a compression and inverse problem task: the network is trained to predict the initial conditions of the dynamics model we used in the simulation, given a target point in space. We investigate this as a regression task, with error propagation in consideration. For target points, up to a radius of 2 kilometers, the model is able to accurately predict the initial conditions of the trajectories, with sub-meter deviation. This simulation-based training process and novel real-world evaluation method is capable of computing trajectories of arbitrary dimensions.